AI build-out reaches far beyond tech

Dave Harrison Smith, CFA
Chief Investment Officer
July 21, 2026

 

The scale of the artificial intelligence (AI) infrastructure build-out is hard to overstate. Pew Research estimated that there were more than 1,500 data center projects planned or under construction in the U.S. as of early 2026. For perspective, Synergy Research Group counted just 659 hyperscale data centers in operation worldwide as of 2021. Capital spending from major cloud and data center providers, including Google, Amazon, Meta, Microsoft, and Oracle, has soared from $130 billion in 2021 to an estimated $760 billion in 2026. That number is expected to increase again in 2027 to nearly $1 trillion.

Capital spending by major technology companies is accelerating
Actual and estimated annual capital expenditures, $billions

While media and investors rightfully focus on semiconductors and suppliers of critical components, the economic impact of this spending extends far beyond these suppliers. The Census Bureau’s Value of Construction Put in Place survey illustrates the breadth of this impact. Construction spending on Data Centers and Electric Infrastructure has risen to more than a quarter of private nonresidential construction spending, up from 17.3% in 2019. Excluding these categories, inflation-adjusted construction spending is up only 1% over the same period. Construction firms, industrial design firms, and a deeply intertwined supply chain of vendors and their employees are all highly levered to the AI infrastructure boom.

Data centers and electric infrastructure now represent more than a quarter of private nonresidential construction
Share of private nonresidential construction spending

 

Rapid revenue growth is supporting exceptional investment

This increase in spending is unparalleled, and investors have rightfully questioned whether the underlying business models can sustain it. Critically, we are seeing exceptional revenue growth at most major AI laboratories. OpenAI is rumored to be targeting annualized revenue growth to $30 billion by year-end 2026, up an impressive 50% from levels at the end of 2025. This is dwarfed by Anthropic’s reported run rate of $69 billion as of mid-July, up from a rumored $9 billion at the end of 2025. For perspective, it took both Google and Facebook approximately twelve years from their founding to approach $30 billion in annual revenue.

AI lab revenue has accelerated sharply
Reported or estimated annualized revenue run rate, $ billions

For now, the rapid revenue growth provides meaningful support for the AI infrastructure build-out. The figures suggest that users are seeing value in AI and that many customers are willing to pay considerable sums to the leading AI labs. Usage, enterprise adoption, and revenue are all rising. Much therefore depends on continued growth across a narrow segment of the economy and a small group of companies.

Yet, disruption is the norm rather than the exception in the technology industry. Internally, we are spending considerable time understanding the impact of the rise of low-cost, open-source models, particularly those emerging from China. DeepSeek, Alibaba’s Qwen, Moonshot’s Kimi, and Zhipu’s GLM have all demonstrated considerable capabilities that may rival some leading AI lab models, at least in some use cases. The ability for corporations to utilize on-premise hardware for these models may help reduce data governance risks. Adoption and further progress here could threaten the economics and business models of the leading labs, pressure pricing, and eventually disrupt the capital-spending forecasts with significant downstream ramifications for the global economy.

 

Competition could reshape the economics

In January 2025, I was in Florida preparing to speak to a group of investors on the unprecedented rise of AI and the emerging investment opportunities. The night before my talk, the Chinese firm DeepSeek shocked investors by announcing the release of a competitive reasoning model at a fraction of the cost of leading Western alternatives. My talk thus had some unexpected fireworks: AI-exposed companies plunged, with Nvidia itself falling 17% in a single trading session.

The DeepSeek moment proved temporary. Revenue growth at the leading Western labs continued to soar, driven by new coding and agentic products. Infrastructure spending moved higher. Lower costs may even have stimulated additional consumption, consistent with a phenomenon in economics and technology called Jevons Paradox.

That outcome is encouraging, but it is no reason for complacency. The impact of significant spending optimization or a pullback in infrastructure spending would reverberate well beyond the tech sector. Near-term evidence remains strong, with usage, revenue, and investment continuing to rise. Given the industry’s dependence on a small group of companies, however, investors should monitor competitive shifts closely. The opportunity remains compelling, but discipline and selectivity are increasingly important.

 

 

# # #